Culture
We strive to build an interdisciplinary team working at the domain of deep learning, NLP, and computational neuroscience. Our ultimate goal is to model humans. Our core values are:
- Experimental: We value scientific rigor, focusing on researching under strong scientific grounds and conducting sound experiments that provide definitive and repeatable findings.
- Computational: Our scientific nature is to use algorithms, mathematical models, strong theoretical background, and strong coding skills.
Lab Entry
We welcomed students from all disciplines but those with a huge passion for research. All Master/Ph.D. students are required to publish at top-tier conferences/journals: HCI (CHI, UIST), NLP (ACL, EMNLP), CV (CVPR, ICCV), Brain (Journal of Neural Engineering).
Current Projects
- Large language models - contributes to the research on how to utilize and improve the efficiency and reasoning/factuality capacity of large language models.
- Multimodal models - contributes to the relationship modeling between vision and text.
- BCI speller - contributes to the development of BCI speller using EEG paradigms such as P300, SSVEP, Hybrid P300-SSVEP and motor imagery for locked-in patients.
- Medical (visual) question answering - contributes to the development of models that generate answers to medical (visual) question answering tasks.
- Legal question answering - contributes to the development of models that generate answers to legal questions, which can help average people access to their rights and laws.
- Informal-formal paraphraser - contributes to the development of models that can help turn informal text into formal text
- Blood glucose monitoring - contributes to the use of Raman spectroscopy and the development of Raman wearables to monitor blood glucose in real-time
Focus Area
Although these topics are slightly different, our lab views them from these shared research challenges:
- Few-shot / semi-supervised learning - contributes to the development of a model that can learn quickly, and when labels are limited or even not available.
- Low compute and efficiency - contributes to solving the problem of efficiency of large model, either by distillation, pruning or quantization or parameter efficient tuning or token merging.
- Robustness - contributes to the use of adversarial or data augmentation for robust performance
- Explainable AI - contributes to the development of tools/techniques or analysis of model that yield better understanding